FAQ & Methodology

How does compatibility checking work?

Every time you add or swap a part in the builder, BlueprintPC runs your build through a set of rules that compare real specs between components — the same checks an experienced builder would do by hand, run instantly. Each rule produces a verdict:

What compatibility checks does BlueprintPC run?

The rules engine cross-checks pairs of components in your build:

• CPU ↔ motherboard socket — the CPU's socket must match the board's (e.g. AM5, LGA1851). A mismatch is incompatible.
• RAM ↔ motherboard — RAM type (DDR4/DDR5) must be on the board's supported list (incompatible if not); RAM speed above the board's rated maximum is a warning, since it will simply run at the board's max speed instead of failing outright.
• PSU wattage — BlueprintPC estimates your system's total draw (CPU + GPU TDP, plus a fixed allowance for the board, RAM, storage, and fans) and compares it to your PSU's rated wattage. Under the estimated draw is incompatible; less than roughly 10% headroom over it is a warning, since PSUs run best with some margin.
• GPU length ↔ case clearance — a GPU longer than the case's maximum supported length is incompatible.
• CPU cooler height ↔ case clearance — a cooler taller than the case's maximum clearance is incompatible.
• Motherboard form factor ↔ case — the board's form factor (ATX, Micro-ATX, Mini-ITX, etc.) must be on the case's supported list.
• CPU socket ↔ cooler — the cooler must list support for the CPU's socket.
• BIOS readiness — some current-generation AMD chipset boards may need a BIOS update to support the newest Ryzen CPUs out of the box; BlueprintPC flags this as a warning so you can confirm firmware support before buying.
• Multi-socket / multi-GPU builds — workstation builds with more than one CPU are checked against the board's socket count and must share a common socket; builds with more than two GPUs get a warning to double-check slot clearance, case space, and PSU capacity.

When a rule fails, BlueprintPC suggests a concrete fix — a different CPU, a compatible PSU wattage, and so on — rather than just telling you something's wrong. Warnings never block a build; they're advisory context so you can decide with full information.

How are performance scores estimated?

Once your build has a CPU and GPU, BlueprintPC shows three estimated performance indices on a 0–100 scale, one per workload:

Each score blends normalized component indices with different weights per workload: Gaming leans heavily on your best GPU with a smaller contribution from single-core CPU performance; Productivity leans on multi-core CPU throughput plus RAM and storage; Content Creation blends multi-core CPU, GPU, and storage, with a bonus for multi-GPU rigs. Raw scores climb roughly 1:1 up to a soft ceiling, then compress gradually toward 100 — so even a flagship single-GPU build typically lands in the low-90s rather than pinning at the top, and multi-GPU or multi-CPU workstation gains stay visible instead of being clipped.

What's the difference between known and estimated scores?

Popular CPUs and GPUs have curated, benchmark-anchored scores ("known"). Parts outside that curated list still get a score — derived from their published specs (clock speed, core count, TDP, VRAM, memory type) using a rough formula calibrated against the known parts — but it's flagged "estimated"since there's no clean spec-to-performance mapping for every part, especially GPUs across different architectures.

What is bottleneck analysis?

BlueprintPC compares your primary CPU's single-core index against your best GPU's index. A large gap either way surfaces a bottleneck callout: a GPU well ahead of the CPU means the CPU is holding back gaming framerates; a CPU well ahead of the GPU means the GPU is the limiting factor. Smaller gaps still surface a lighter, "medium"severity note at high framerates.

These are estimates for comparing builds against each other, not lab-measured frame rates — actual performance varies by game, application, drivers, cooling, and the rest of your system.

What is the BlueprintPC Bench leaderboard?

The Benchpage is different from the builder's estimated scores — it's built from real hardware runs. BlueprintPC Bench is a small local client you download and run on your own machine; it measures CPU single-thread and multi-thread throughput, memory bandwidth and latency, disk sequential/random read-write speed, and GPU compute, then lets you opt in to share the anonymized results to the public leaderboard.

How is the benchmark index calculated?

Every raw measurement is converted into an index anchored to a fixed, documented reference system — a reference index of exactly 100. Faster-than-reference hardware is uncapped and can score above 100; slower hardware scores proportionally below it. For metrics where lower is better (like memory latency), the formula inverts so a lower raw number still produces a higher index. Per-metric indices are then combined into the same three workload composites (Gaming, Productivity, Content) using workload-specific weights — gaming leans on GPU compute and single-thread CPU; productivity leans on multi-thread CPU and memory bandwidth; content leans on multi-thread CPU, GPU compute, and disk throughput.

Is my benchmark data anonymous?

Sharing a run to the leaderboard is entirely optional and anonymous — no account or personal information is attached to a submission. Every result the client sends is cryptographically signed before it leaves your machine, and BlueprintPC's server re-verifies that signature before a run is ever accepted or displayed, so scores can't be forged or edited in transit.

How are prices and “Best Value” determined?

Where available, BlueprintPC shows live prices pulled from Canadian and US retailers, so you can compare a build's total cost across sources instead of guessing. When you save multiple builds to compare, the one with the lowest total price across its parts is marked Best Value; the one with the highest estimated performance is marked separately as Best Performance— they're not always the same build.

Not every part has live pricing at all times; when a component is missing a price, a build's total shows as a starting-from amount instead of a hard number.

Some buy links are affiliate links to retailer partners — BlueprintPC may earn a small commission if you buy through one, at no extra cost to you. Every affiliate link carries a visible Ad disclosure, and affiliate relationships never influence a compatibility verdict, performance score, or which part is recommended. Full details live on the Affiliate Disclosure page.

How we estimate game FPS

Every game page and the “Will it run?” calculator show a framerate — but none of it comes from us running the game on your exact hardware. Here is exactly where the numbers come from and what they do and don't mean.

Per-game calibration anchors

For each game, we start from a small set of calibration anchors — real framerates at a given resolution, sourced from published review benchmarks run on real reference hardware. That anchor is the ground truth for the whole estimate; everything else is a scaling calculation from it.

Scaled by GPU index, capped by a CPU ceiling

To project FPS for a GPU other than the reference card, we scale the anchor by that GPU's index relative to the reference GPU's index. That scaled number is then checked against a CPU ceiling derived from the CPU's single-core index and the game's own CPU-heaviness rating — a fast GPU paired with a weak CPU gets capped at what the CPU can actually feed it, the same bottleneck logic used in the builder.

Every figure carries a visible Estimated label. These numbers are a comparison tool for shortlisting hardware against a specific game — not a lab measurement of your exact system, driver version, and settings. Actual in-game framerate varies with your specific settings, resolution scaling, background load, and game patches after our anchor was taken.

How the VRAM fit math works

The AI model-fit calculator and every AI-vertical page answer one question — does this model fit on this GPU — by adding up three real costs and comparing the total to a card's VRAM.

The three components of VRAM need

• Model weights — the published GGUF file size at your chosen quantization (Q4, Q5, Q8, FP16, etc.) when a model publishes one; otherwise estimated from parameter count and bytes-per-weight at that quant level.
• KV cache — the working memory the model needs to hold context as it generates. This grows with your context-length setting, which is why a model that “just fits” at a short context can run out of room on a long document.
• Overhead — a roughly 15–20% allowance for the inference runtime itself, CUDA/ROCm buffers, and other fixed costs that aren't the model or its cache.

Tokens-per-second estimate

Where we show an expected generation speed, it's approximated as the GPU's memory bandwidth divided by the model's active weight size — the standard back-of-envelope method for memory-bandwidth-bound decode, and the same approach used to calibrate unified-memory platforms like Apple Silicon and Strix Halo against discrete GPUs. It's always shown with an Estimated label — real throughput depends on your specific inference engine, batch size, and prompt length, and no community-scale consumer token-per-second database exists yet for us to calibrate against directly.

Still stuck?

If something in your build still doesn't add up, we're happy to look at it.